Pika's MCP gains MiniMax H3 video generation
By Steven Van ·
Developers can now call H3's Omni-Reference, 2K, native-audio output through Pika's MCP server, the same day it landed on Krea, Magnific, Leonardo, Lovart
Pika announced that MiniMax H3 is now available on the Pika MCP, giving developers and AI agents direct access to MiniMax's new video model through Pika's Model Context Protocol server rather than only through the Pika app itself.
Why the MCP matters here
Pika's MCP lets tools like Claude Code, Cursor, or any other MCP-compatible client call Pika's video models programmatically, generating video as part of an agentic workflow instead of a manual upload-and-prompt session in a browser. Adding H3 to that surface means the model's Omni-Reference control, up to 9 reference images, 3 video clips, and 3 audio clips, alongside 2K, native-audio output, becomes something an agent can call directly when building or automating a content pipeline.
Following Seedance's playbook
Pika had already put Seedance 2.0 in native 4K on the MCP; H3 follows the same pattern, keeping the MCP current with whichever frontier video model has the best combination of quality and cost at any given moment, rather than locking developers into a single provider.
Part of a coordinated rollout
H3 reached the Pika MCP the same day it launched on Krea, Magnific, Leonardo, Lovart, and Runway, underscoring how fast a strong new model now propagates across the AI video tooling ecosystem.
Sources: Pika on X.